Skip to content

Stela v0.4.2

Latest

Choose a tag to compare

@github-actions github-actions released this 15 Nov 05:13
· 321 commits to main since this release

Install

Download the archive for your system, unpack it and plug in the USB stick. Then run one of these in the unpacked folder.

Make a new stick (this erases it):

sudo ./stela

Update an existing Stela stick to v0.4.2:

sudo ./stela update

On Windows, open Terminal with Run as administrator and write .\stela.exe in place of sudo ./stela.

On macOS, stela is not notarized. If macOS says it cannot be opened, run this once in the unpacked folder:

xattr -d com.apple.quarantine stela

Then boot the PC from the stick.

Changes

Boot

  • [#80, #89, #90, 1127, 1048] Intel: the first start of macOS after an install no longer stops with native continuation admission: CorruptContext. A page table root loaded under a PCID was refused.
  • [#78, #76, #91] AMD without IA32_TSC_ADJUST: a processor's TSC is no longer written. Each processor keeps a software offset instead (Non-monotonic time panic). This replaces the 0.4.0 fix.
  • [#76, #78] A device clock that cannot reach the board's count stops the Mac with that reason (SPMI Read RTC alarm panic before). Why some boards hand over such a count is not known yet.
  • [#87, #105] A PC with fewer than 8 processor threads no longer panics in setCPUCoreEnergyAccumulator or enableCPUComplex. One processor of each cluster is kept. A PC with one usable thread is refused with the reason.
  • [#74] A PC with no UART at COM1 no longer feeds the port's reads to the macOS console. The automatic Metal plugin repair was interrupted by them. stela repair asks for the repair again.
  • [#84, #88] Laptops: the PCI inventory passes over functions that are absent (vendor ID 0000, an idle CardBus header). It refused the GPU before.
  • Firmware AML that reads PCI configuration space through memory is served. The DSDT failed to load on some Intel boards.
  • [#102] A PCI header read of a bus that no root bridge forwards and no MCFG window holds answers all ones. A host function with a bridge header and a hard-wired Bus Master bit is accepted as a host bridge's is.
  • [#81] The engine heap grows with the PC's memory, up to 4 GiB. A 16 GiB PC failed a 17 MB allocation at the desktop. An allocation failure names the heap's live, largest free and total bytes.
  • The other processors start under the memory types the boot processor holds when the firmware changed them late. Every refused processor start names its reason.
  • A processor that crawls inside the macOS kernel no longer has its clock republished at each short stop (mutex interlock timeout panic). A host step over 250 ms is logged with its part.
  • A damaged ACPI table that Stela does not need no longer stops the start. MADT defects that Linux tolerates are tolerated.
  • Stela ends the firmware's boot services itself, with a bounded retry.
  • CR4.MCE is set on every processor. A machine check is reported with its banks.
  • Processors that run no part of the Mac are parked in an idle loop.
  • One processor whose firmware check fails, or that misses a step of the start, is left out with a line. The others are used.
  • The other processors start from a boot services page below 1 MiB when no free page is left there.
  • xAPIC is kept when the firmware's DMAR sets the x2APIC opt-out.
  • A bridge with stale or crossing bus numbers is left undriven with the buses behind it.
  • A stop before the heap exists is kept in firmware variables and shown by stela log.
  • A stick made new on a PC that already holds a Stela macOS offers that macOS when the stick has no disk record. The menu offered only Install macOS before.
  • An installed macOS starts as the machine it was installed as (its ECID), which its machine state and its boot ticket name. A stick made again has another ECID, and macOS panicked at seputil when started with it.
  • A PC with a repair record from 0.4.1 runs the automatic Metal plugin repair once more.
  • [#94] A refused MMIO aperture names the check that refused it.
  • Every stop shows its reason on screen.
  • A step that makes no progress for its bound ends in a named stop.
  • AMD: the cause of the last reset is logged.

Disks and USB

  • [#77] Intel VMD: BARs of mixed width are decoded, two NVMe drives behind one VMD are handled, and an NVMe behind VMD works with no IOMMU. The same changes were tested by Fisherman6a in #77.
  • [#75] NVMe: a completion is taken by its command identifier, as Linux does.
  • [#92] SATA: internal commands get Linux's timeouts, a failed first IDENTIFY is retried after a hard reset, and link power saving is refused during the first reset. An expired command is named with the port's registers.
  • NVMe: the quirks Linux keeps per drive are applied (power states, Write Zeroes, ready delay, namespace scan).
  • NVMe: bus mastering is enabled only after the controller the firmware left running has stopped.
  • A keyboard, mouse or hub plugged in after boot is started in steps. Devices behind a KVM switch, dock or monitor hub come back after a switch. Keys held at an unplug are released.
  • Other USB devices plugged in after boot are reported as unused until the next boot.
  • A USB storage device gets its first class request one second after its configuration, as Linux does, on xHCI and EHCI.
  • A USB stick that does not answer at the start is reset and started again.
  • Install macOS's disk picker marks a spinning disk as maybe too slow for macOS.
  • [#109] A USB stick with no write cache is sent no SYNCHRONIZE CACHE, as Linux does, and a stick that refuses the command is sent no more. The translation and pipeline caches were never written to such a stick.
  • A USB stick Stela uses that leaves the bus after boot and comes back at the same port, with the same identity and the same first two blocks, is started again. Its log and stores go on.
  • [#96] Install macOS's disk picker reads a SATA or IDE disk's serial through the firmware's ATA pass-through too. A disk without a serial logs what the firmware offers for it.
  • A USB Ethernet adapter that is unplugged or stops answering no longer holds a service turn.
  • Realtek Ethernet boards sold under D-Link, Allied Telesyn, US Robotics and NCube IDs are driven, as Linux's r8169 does.
  • The screen names each Ethernet adapter that is not driven, with its IDs, also when another adapter started.
  • [#88, #93, #99] The boot log on the stick no longer holds other disk work while it writes. A store of 13 to 58 s held a processor and macOS panicked in its watchdog.
  • The two NVRAM banks are stored one after the other. A stuck disk no longer holds a restart or power-off.

Graphics and display

  • [#83, 1110] Hybrid laptops: an NVIDIA card whose registers did not answer is brought to D0 and has its memory decoding enabled (the GPU's register BAR does not decode).
  • [#86, #100] GTX 16 and RTX 20: the ACR bootloader file is on the stick (fwsec-sb failed with -2). Contributed by cn0xroot in #98.
  • [#86] A fault on a GPU client mapping that moved is served in the faulting context (InterruptContext stop).
  • [#85] A failed GPU start ends in a named stop with the driver's first error, not vector=6.
  • [#85] RTX 5060 and 5060 Ti: the stop names the GPU firmware's own result (GSP-RM 570.144 ends its start with a timeout). These cards do not start yet.
  • A failed buffer or heap allocation in the Metal plugin is logged.
  • [#75] A read of a texture's bytes goes through a GPU copy into cached memory. Reads from video memory made Setup stutter.
  • [#75] A depth or stencil texture that asks for shader writes is made without them, as on a Mac.
  • Safari and other WebKit views no longer draw pages as horizontal stripes. macOS is told once that the GPU has no universal buffer compression (the disable-buffer-compression preference, set by a one-time job unless a value is already there).
  • Display sleep turns the monitor off. A wake lights it again.
  • A monitor that comes back is set in a mode from its new list. One refused output no longer stops the others.
  • Every connected output that runs a shared size shows the picture.
  • A 32:9 monitor runs the largest mode of its own shape.
  • The GPU is unloaded in order at power-off. A GPU an earlier run left running is reset through its bus.
  • A GPU whose firmware files are not all on the stick is refused with their names.
  • A GPU call that no task takes within 300 ms is answered with a timeout.
  • [#81] GTX 16 to RTX 40 on DisplayPort: when a monitor's EDID read gives nothing, it is read again with each I2C read ended by its last chunk, as on RTX 50. A monitor's EDID arrived shifted and was refused.
  • [#107] A GPU that is refused before its start shows the reason on screen.
  • [#77] A boot processor whose timer interrupt does not arrive during the GPU start ends in a named stop after 2 s. The screen stayed black before.

Input and power

  • Keys and buttons are kept per device. A touchpad report no longer ends a mouse drag.
  • A power button press is a normal press for macOS. A second press within 3 s forces the PC off.

Install

  • [#77] An online install's download resumes inside a file after a network wait. It restarted the file before. As in the patch by Fisherman6a.
  • [#93, #99] Reads of the install stick are bounded to 20 ms in each service turn, all together, and go on at the next. A read behind a slow write of the stick held a processor for 20 to 30 s. A file read in many pieces resumes where it stopped.
  • The disk picker refuses disks that report one serial number twice.
  • [#108] An online install's network reads are bounded to 20 ms in each service turn, all together. A processor was held for seconds.
  • [#106] A stop whose record the firmware's variable service fails to save names that failure and the reason before it.

CLI

  • Plain stela runs every choice to its end for every stick state. --yes runs what is due without a terminal.
  • A question that erases or downloads takes Quit on Enter, and names what is on the disk.
  • stela create /dev/rdiskN works on macOS.
  • stela repair puts back a missing stela.efi and old boot files.
  • stela log says when the stick's own boot log was not found or could not be read, and why. A log file with a damaged header is saved with the lines it holds.
  • EFI partitions that a run mounted are unmounted on every way out.
  • A wrong computer clock is named when the server's certificate is not valid yet or has expired.
  • A declined question ends with "Nothing was changed" and exit 0.
  • A card reader with no card is not offered as a stick.
  • [#106] stela status and stela repair mount the stick's data partition when it is not mounted. An unmounted stick was reported as needing repair.
  • A partition erased by 0.4.1 is blank again for Install macOS: its first MiB is zero and no known file system, volume manager or encryption signature is found at the offsets checked.

Wi-Fi and Bluetooth do not work in this release.

Boot log

If Stela does not work, save the boot log and include log.txt in your report:

sudo ./stela log --output log.txt